|
(11) | EP 0 444 885 A2 |
| (12) | EUROPEAN PATENT APPLICATION |
|
|
|
|
|||||||||||||||||||||||||||
| (54) | A method for controlling a buffer memory for a magnetic disk storage system |
| (57) The present invention provides a method for temporarily storing and retrieving 8-bit
character information data for a magnetic disk information storage system in a number
of 4xn DRAM buffer memory configurations, characterised by the steps of: providing
a virtual memory address for each item of 8-bit character information data, said character
information data being organised into a 16-byte block, translating the virtual memory
addresses into corresponding addresses of memory locations in said 4xn buffer memory
for storage of 4-bit groups of said 16-byte block in said 4xn DRAM buffer memory,
said translating step including: selecting a row address for storage of said 16-byte
block, selecting a base column address for said 16-byte block, and successively incrementing
said base column address to provide additional column addresses for successive 4-bit
groups of said 16-byte block, and transferring each of said 4-bit groups of said 16-byte
block through a 4-bit data bus to the various pre-determined address locations in
one of said 4xn DRAM buffer memory configurations determined by said translating step.
In addition, and 8-bit parity word for the 16-byte block is stored in a separate part
of the DRAM buffer memory. |
[0001] This invention relates to a method for controlling buffer memories for magnetic disk storage systems.
[0002] Temporary storage devices such as dynamic random access memory (DRAM) devices are often used in connection with magnetic storage media as buffer memories for temporarily storing character information. Each character is represented, for example, by an 8-bit information word and 1-parity bit. Typically, information is read into a temporary storage memory device one character at a time using nine parallel data bit lines. DRAMs are typically available as 4- or 1-bit wide devices, so that storage of the extra parity bit associated with each information word is cumbersome, resulting in the parity bit not being stored or being stored in a separate memory device.
[0003] It is an object of the invention to provide a controller for and a method of controlling DRAM buffer memories for a magnetic disk storage system, which permits efficient use of standard low cost DRAM memories for storing information data and parity data in a variety of DRAM memory modes.
[0004] In accordance with the present invention, there is provided a method for temporarily storing and retrieving 8-bit character information data for a magnetic disk information storage system in a number of 4xn DRAM buffer memory configurations, characterised by the steps of: providing a virtual memory address for each item of 8-bit character information data, said character information data being organised into a 16-byte block, translating the virtual memory addresses into corresponding addresses of memory locations in said 4xn buffer memory for storage of 4-bit groups of said 16-byte block in said 4xn DRAM buffer memory, said translating step including: selecting a row address for storage of said 16-byte block, selecting a base column address for said 16-byte block, and successively incrementing said base column address to provide additional column addresses for successive 4-bit groups of said 16-byte block, and transferring each of said 4-bit groups of said 16-byte block through a 4-bit data bus to the various pre-determined address locations in one of said 4xn DRAM buffer memory configurations determined by said translating step.
[0005] Advantageously, an error detection code word in the form of an 8-bit parity word is provided for the 16-byte block and is stored by selecting a parity word row address, selecting a parity word base column address for a first 4-bit group of said 8-bit parity word, and incrementing said parity word column address to provide a second column address for a second 4-bit group of said 8-bit parity word. The error detection code word is then transferred through the 4-bit data bus to the various pre-determined address locations in the 4xn buffer memory for said 8-bit parity word.
[0006] The translating step may include selecting row and column addresses using multiplexer circuits. The method may further include the step of latching the row and column addresses into latch circuits, the incrementing step including the step of feeding the outputs of the latch circuits through multiplexing circuits connected to the inputs of the latch circuits to increment the column addresses.
[0007] Optionally, the 4xn buffer memory may comprise a DRAM operating in a page mode or in a static column mode.
[0008] The invention is described further, by way of example, with reference to the accompanying drawings, in which:-
Figure 1 is an overall functional block diagram of an integrated circuit SCSI controller which includes a DRAM interface circuit for communicating with a DRAM buffer memory over a 4-bit data line;
Figure 2 is a block diagram of a buffer memory control portion of the SCSI controller;
Figure 3 is a block diagram of the buffer memory control portion showing the interfaces required to transfer data from a disk memory to a host computer;
Figure 4 is a block diagram showing storage of character information data and parity data in the buffer memory when data is written from a host computer to a disk memory;
Figure 5 is a timing diagram for a normal page-mode block transfer of data between the controller and a DRAM buffer memory;
Figure 6 is a timing diagram for a static-column-mode block transfer of data between the controller and a DRAM buffer memory;
Figure 7 is a timing diagram for an inter-leaved page-mode block transfer of data between the controller and a DRAM buffer memory;
Figure 8 is a block diagram of the DRAM interface circuit of the controller;
Figure 9 is a circuit diagram of a state machine portion of the DRAM interface circuit; and
Figure 10 is a circuit diagram of an address sequencer portion of the DRAM interface circuit.
[0009] Figure 1 shows a block diagram of a 68-pin integrated circuit small computer system interface (SCSI) controller circuit 10. As shown in the diagram, various interfaces are connected by signal buses to a memory control unit 12. A formatter 14 provides an interface to a magnetic storage disk through various signal lines. An NRZ signal on a line 16 is an input/ output signal in an NRZ data format, which provides a data bit stream to or from the logic in the formatter 14. An RD/REF CLK signal on a line 18 is a read/ reference clock, which is the clock signal for the formatter 14 supplied by a data separator of the magnetic disk electronics. The frequency of this clock signal ranges from 5 Mhz to 24 Mhz. A read gate signal RG on a signal line 20 is a signal which enables a read channel and causes the controller circuit 10 to input NRZ data from the magnetic disk. A write gate signal WG on a signal line 22 is a signal which enables write drivers and causes the controller circuit 10 to output NRZ data to the magnetic disk. An INPUT/COAST signal on a line 24 and an OUTPUT signal on a line 26 are general purpose signals used to synchronise the formatter 14 with external hardware. An INDEX signal on a line 28 is an index signal from the magnetic disk drive and is supplied once per revolution of the disk. A SECTOR/WAM/AMD signal on a line 30 is a sector input, address mark detected input, or write address mark output to or from the formatter 14, depending on the operating mode of the system. The formatter 14 also contains an error correction code circuit ECC 32. Signals passing between the memory control unit 12 and the formatter 14 are carried on a signal bus 34.
[0010] Communication between the memory control unit 12 and an external micro-processor, such as an 8051 Intel Controller, is handled with a micro-processor interface circuit 40, which communicates with the memory control circuit 12 through a bus 42. An active low chip select signal CS on a signal line 44 enables the controller integrated circuit 10 for either a read or a write operation. An active low read data signal RD on a signal line 46 in conjunction with the CS signal causes data from a specified register within the memory control circuit 12 to be moved to a data bus AD provided by a bus 52, as indicated. The micro-processor address/data bus AD (7:0) is an input/output bus with active high signals provided on bi-directional signal lines, which interface with a multiplexed micro-processor address/ data bus of the external micro-processor. An active low write data signal WR on a signal line 48 in connection with the CS signal causes data from the data bus AD to be moved to a specified register within the memory control unit 12. An interrupt request signal IRQ on a signal line 54 is an active low output signal from a micro-processor interface interrupt control circuit 56 to interrupt the external micro-processor. A clock control circuit 58 within the integrated circuit 10 provides appropriate clocks to a clock bus 60 to the micro-processor interface circuit 40. Signals from the micro-processor interface circuit 40 to the formatter 14 are provided on a bus 62.
[0011] An SCSI interface circuit 70 provides for communication on an SCSI bus 92 from a host computer to the memory control unit 12. Active low input/output signal lines DB (7:0) are the SCSI data lines. An active low input/output busy SCSI control signal BSY is provided on a signal line 74. An active low input/output SCSI SEL control signal is provided on a signal line 76. An active low input/output command/data SCSI control signal C/D is provided on a signal line 78. An active low input/output message SCSI control signal MSG is provided on a signal line 82. An active low input/output request signal REQ in connection with an active low input/output acknowledged signal ACK on a line 86 forms an SCSI data transfer hand shake. An active low input/output SCSI attention control signal ATN is provided on a signal line 88. An active low input/output SCSI re-set signal is provided on a signal line 90. All of the above SCSI control signals are provided in accordance with the SCSI standards. The SCSI interface circuit 70 is coupled to the memory control circuit 12 through the signal bus 92.
[0012] A DRAM interface circuit 100 for providing communication with a DRAM buffer memory is coupled to the memory control circuit 12 through a signal bus 101. External connections from the DRAM interface circuit 100 to the DRAM buffer memory are provided through various signal lines. An active low output row address strobe signal RAS on a signal line 102 provides an address strobe for the DRAM buffer memory. An active low input/output column address strobe signal CAS on a signal line 104 provides an address strobe for a first or an only DRAM. A second active low output column address strobe signal CAS2 on a signal line 106 provides an address strobe for a second DRAM. An active low output write strobe signal W is provided on a signal line 108. An active low output enable signal G on a signal line 110 is a DRAM output driver enable signal. An address bus 112 provides nine bits of DRAM address A (8:0). Active high input/output signals are provided on a DRAM data bus DQ (3:0).
[0013] Figure 2 shows diagrammatically the interface circuits for the integrated circuit 10. The data is transferred to and from the three buses 72, 34, 52 and the DRAM data bus 114. Data from the SCSI data bus 72, that is host data, is input to a host FIFO 120. Formatted data from the formatter 14 is input to a format FIFO 122. Read/write data from the microprocessor on the bus 52 is input to a window RAM 124. Byte/wide data transfers to and from each of these asynchronous interfaces are made to the respective FIFOs. Nibble/wide transfers are made to and from the DRAM between the respective FIFOs under the control of a state machine 126.
[0014] Several different DRAM configurations can be implemented, which include: one 64K x 4 DRAM page mode or static column mode; two 64K x 4 DRAMs; one 256K x 4 DRAM page mode or static column mode; and two 256K x 4 DRAMs. The circuit also provides two nibbles of parity for each thirty two nibble transfer. Transfers to and from the DRAM are in blocks of thirty two nibbles, or 16-bytes. An arbitration circuit 128 gives priority for each interface operation with the DRAM. The disk sequencer for the magnetic disk which provides formatted data from the format FIFO 122 has the highest priority. Refresh of the DRAM buffer memory has the next highest priority. Data from the micro-processor from the read/write bus, which is stored in the window ram 124, has the next highest priority. Finally, host data on the SCSI bus 72 stored in the host FIFO 120 has the lowest priority. The arbitration circuit 128 arbitrates at the end of each 16-byte transfer and allows only one interface access to the DRAM at a time. A parity generator/checker circuit 130 provides the two nibbles of parity for each thirty two nibble transfer.
[0015] Figure 3 is a block diagram of the buffer memory controller showing the interfaces required to transfer data from a disk memory to a host computer on the SCSI bus. The controller can handle three asynchronous interfaces to the DRAM buffer memory 130. Figure 3 diagrammatically shows various sectors of the DRAM buffer memory 130 being utilised. As the disk information is filling sector n+2 of the DRAM buffer memory 130 from the disk or format FIFO 122, the microprocessor can be correcting sector n+1, and the host computer connected to the SCSI bus 72 can be reading sector n through the host FIFO 120.
[0016] Figure 4 diagrammatically shows the storage character information data and parity data in the buffer memory when the data is being written from the SCSI bus 72 to the disk memory connected to the bus 34. Information from the host FIFO 120 is input to the DRAM in 16-byte blocks of data. Parity for that block is generated by the parity generator 130a and stored in the parity portion of the DRAM buffer memory 130. For read out, the data is read out from both the parity storage area and the information storage area of the DRAM buffer memory 130 to the form FIFO 122. Parity checking on the output data being read out from the DRAM buffer memory 130 is performed by the parity checker 130b. Parity is generated during the write mode and it is checked during the read mode. Parity is written in a separate place in the DRAM with 2K bytes of address space required for parity for every 32K bytes of total memory.
[0017] Figures 5, 6 and 7 show timing diagrams for various DRAM buffer memory configurations to and from which blocks of data are transferred between the memory control logic 12 and the DRAM buffer memory 130. The DRAM interface 100 operates on blocks of data, which are 16-bytes in length. All transfers of lesser amounts of data are padded by the control logic prior to the DRAM interface 100 receiving the blocks of information. The buffer control logic also calculates parity, which is two nibbles of longitudinal parity for thirty two nibbles of data. The DRAM interface 100 performs a remapping of the base address of the information to simplify the internal logic. However, parity is always placed in the top 1/16th of the address space of the DRAM buffer memory as seen from inside. Various sizes of DRAM buffer memory can be used and these sizes include: 32K, 64K, 128K and 256K. One 256K DRAM provides a 32K buffer memory. Two 256K DRAMs provide a 64K buffer memory. One 1M DRAM provides a 128K buffer memory. Two 1M DRAMs provide a 256K buffer memory.
[0018] The transfer sequence for a block of data to a normal page mode DRAM is shown in Figure 5. First, the row address is presented. Then thirty two columns of data are transferred. Then, the parity row address is presented followed by the parity information from the two parity columns. Note that if a single page mode DRAM is used, each column of information except the last is followed by a column pre-charge cycle.
[0019] Figure 6 shows the timing signals for a single static-column mode DRAM. If a single static-column mode DRAM is used, the information from a column is transferred one after another with no intervening precharge cycle required. This greatly increases the speed of data transfer. If two DRAM memories are used, they are auto-matically alternated so that a pre-charge cycle for one DRAM occurs at the same time as the column cycle for the other. This results in the same speed as a static-column DRAM would give. Thus there is no advantage in using static-column DRAMs, which are more expensive, in a two RAM configuration. If it is necessary to use static-column DRAMs in a two RAM configuration, these memories can be used in a page mode of operation.
[0021] From the timing diagrams, it can be seen that transfer of a single 16-byte block of information in a normal page mode of operation takes 72 cycles of the DRAM clock, while the same transfer in static-column or inter-leaved page mode is 40 cycles of the DRAM clock. With this information, the calculation of the required speed of a DRAM to achieve a buffer band width, as well as the speed of the clock, can be made. For a single page mode DRAM, the minimum cycle time required is equal to the column access time. While for a single static-column DRAM or two page mode DRAMs, the minimum cycle time is equal to the column access time plus an address control skew time. It is recommended that the nominal cycle time be set somewhat slower than the minimum cycle time to allow for speed variations due to voltage ripple on the DRAM power supplies. If a 32K buffer using a single Texas Instruments TMS4464-12 256K page mode DRAM is used, a 60 nS column access time is obtained. The clock is set for a cycle time greater than 61.8 nS. The worst case slow cycle time will be 63.8 nS for a total time of 4,594 nS to transfer a 16-byte block of information or a guaranteed band width of 3.48 megabytes/second. If two 256K page mode DRAMs are used, the clock is set to greater than 64.9 nS with a worst case slow time of 66.9 nS, taking 2,676 nS to transfer a block of information with a guaranteed band width of 5.98 megabytes. If a full 8-megabyte band width is required, a block of information must be transferred every 1,961 nS, allowing for refresh, and the worst case time will be 49 nS. This would require the use of a static-column DRAM or a pair of page mode DRAMS with a column access time less than 45.7 nS.
[0022] Figure 8 is a block diagram of a control logic circuit. It shows the main elements of the DRAM interface circuit 100 of Figure 1. The DRAM interface circuit 100 includes a control state machine (Control FSM) 200 and an address generator 202. The address generator 202 provides the address signals on a signal bus 112 to the DRAM buffer memory. A DRAM clock signal is provided at an input terminal 204. The DRAM clock signal is distributed to the control state machine 200, the address generator 202 and a block of four pairs of D flip flops, one of which is typically shown as reference numeral 206. Each pair of the D flip flops processes, respectively, one bit of four bits of read data, received on a signal bus 208, or write data, received on a signal bus 210. Each of the respective read data bits and write data bits are re-timed with the DRAM clock in respective D flip flops 212, 214. The block 206 shows the re-timing circuit for the ith bit. The re-timed read data is re-assembled in a data bus 216 and the re-timed write data is assembled in a data bus 218, as shown.
[0023] The control state machine 200 receives the DRAM clock on a signal line 220. A re-set signal is received on a signal line 222. A start signal is received on a signal line 224. A refresh command signal is received on a signal line 226. A signal indicating that a static column DRAM mode is being used is received on a signal line 230. A signal indicating that two DRAMs are to be used is received on a signal line 232. The control state machine 200 provides an inverted row address select signal RAS on the signal line 102; an inverted column address signal CAS on the signal line 104; a second column address signal CAS2 on the signal line 106; an inverted DRAM write strobe W on the signal line 108 and a DRAM output drive enable signal G on the signal line 110. When 16-bytes of information has been transferred a DONE signal is provided on a signal line 234.
[0024] The address generator 202, as previously mentioned, provides nine address bits on the signal bus 112 to the DRAM. The address generator includes a counter, which counts from the base address provided as an input to the address generator on a signal line 236. A signal indicating that two DRAMS are being used for buffer storage is provided on a signal line 238. A signal indicating whether one megabit RAMS are being used is provided on a signal line 240. The counters within the address generator 202 are enabled by a count enable signal provided from the control state machine 200 on a signal line 242. A signal is output from the address generator 202 on a signal line 243 to the control state machine 200 when a count of thirty two is reached in the address generator indicating that the end of a block of data has been reached.
[0025] Signals pass between the control state machine 200 and the address generator 202 on signal lines 242, 244, 245, 246, 247 and 248. These are steering and control signals for the multiplexer logic. A signal Apply_Row on line 248 steers the multiplexer to form the data row address. A signal Apply_Col on the line 247 tells it to form the data column address. A signal Apply_Par_Row on the line 246 forms the parity row address. A signal Apply_Par_Col on the line 245 forms the parity column address. A signal Apply_Ist_Col on the line 244 says that this is the first column address of either the data or parity group, and causes the count bits to be set to zero. A signal Count Ena on the line 242 causes them to count. If the signal Apply_Col or Apply_Par_Col are true, and neither of the signals Apply_1st_Col nor Count_Ena are true, the count bits will be held as they are. If the signal Terminal_Count on the line 243 is the signal back to the control state machine 200 that the count sequence for the data bits is completed, and allows the control state machine 200 to move onto the states in which parity is written or read.
[0026] Functionally, the DRAM interface circuit shown in Figure 8 corresponds with the DRAM interface circuit 100 shown in Figure 1.
[0027] Ordinarily, to support a large number of different sizes of DRAMs, it would be required that a large number of addresses be multiplexed by a circuit such as the address generator 202. However, it is possible to fold the addresses required in such a way that the number of address bits required to provide all of the addresses required for the memory for a given bit can effectively be reduced from nine to approximately four. Table 1 as shown below shows a conventional scheme for providing connections for fourteen bits of internal addresses, designated as IA14 to IA4 (as provided on the signal bus 236) to the DRAM address lines designated A8 to A0 (provided on the address bus 112) to the DRAM.
[0028] Table 1 is shown for two DRAM buffer memory sizes The first is a 32K DRAM buffer memory size, which is implemented with a 64K x 4 DRAM. The second is a 64K DRAM buffer memory size, which is implemented by two 64K x 4 DRAMs. For each of these cases, various ones of the bit lines for the internal addresses are connected to the DRAM address bit lines through multiplexers. For example, with reference to the timing diagram of Figure 6, the internal addresses for the data row are provided in the first line of the Table, that is in the ROW line. DRAM address line A7 receives the internal address line 14. DRAM address line A6 receives the internal address line 13. DRAM address line A5 receives the internal address line 12 etc. The addresses for the data columns are provided as thirty two different column addresses in the COLUMN line of Table 1. DRAM address bits A4 to A0 are provided by counter bits CNT4 to CNT0, which are the output bits of an internal counter. The third row in Table 1 contains the parity row addresses and the fourth row in Table 1 contains the parity column addresses. For the first example of Table 1, thirty five addresses are provided to the DRAM to access a row, thirty two columns and a parity row and column. Similarly, the second example of Table 1 shows the address multiplexing required for a 64K DRAM buffer memory.
[0029] Note that for Table 1, in order to accommodate two different memory configurations, eight different internal address lines are typically required to be changed. These changes are accomplished by multiplexing the appropriate internal address line and counter output lines to selected DRAM address lines.
[0030] Table 2 shows how the logic according to the invention maps internal addresses and counters to the DRAM address lines for four different memory configurations. Note that the notation IAn refers to the internal address, assuming that this information is presented as bits 17:4, and CNTN refers to the nibble count of 1:32. If two DRAMs are used, only four bits of nibble count are used. Comparison of the first two examples of Table 2 with the examples of Table 1 shows that for Table 1 the DRAM bit line A7 has seven different internal bit lines multiplexed into it. For the first two examples of Table 2, the bit line A7 has only three internal bit lines multiplexed into it. Thus, Table 2 requires a smaller set of internal address multiplexers to map the internal addresses to the DRAM address bits. This provides a significant reduction in circuit complexity.
[0031] Figure 9 is a more detailed block diagram of a control state machine 200, shown in Figure 8. The control state machine 200 includes a first program logic array (PLA0) 300 and a second program logic array (PLA1) 301, which provide combinatorial logic for combining the various input signals to the control state machine 200 to provide logic output signals. The DRAM clock at the terminal 204 clocks a number of D flip flops to latch the combinatorial outputs of the program logic arrays 300, 301. Various other logic elements are provided at the output terminals of the various D flip flops to provide the output signals at the various terminals, as indicated in the drawings.
[0032] Figure 10 shows the circuit details for the address generator 202 shown in Figure 8. Essentially, these circuits implement the connection arrangement of Table 2. The internal address signals provided on the bus 236 are distributed on that bus to various and/or/invert stages, which function as multiplexers 401 to 409 for selection of these address lines in accordance with the internal logic provided by the control state machine 200 and various other logic elements as shown in Figure 10A and 10B. Each of these multiplexers 401 to 409 has its output signal coupled to the D input terminals of a respective D flip flop 410 to 418, which latch the output signals of these multiplexers under control of the DRAM clock provided at the terminal 204. The various column count signals CNTN as indicated in Table 2 are provided by output signals from certain of the D flip flops 413 to 418 being coupled through multiplexers 420 to 425 and latched into the inputs of the D flip flops 413 to 418.
1. A method for temporarily storing and retrieving 8-bit character information data for
a magnetic disk information storage system in a number of 4xn DRAM buffer memory configurations,
characterised by the steps of: providing a virtual memory address for each item of
8-bit character information data, said character information data being organised
into a 16-byte block, translating the virtual memory addresses into corresponding
addresses of memory locations in said 4xn buffer memory for storage of 4-bit groups
of said 16-byte block in said 4xn DRAM buffer memory, said translating step including:
selecting a row address for storage of said 16-byte block, selecting a base column
address for said 16-byte block, and successively incrementing said base column address
to provide additional column addresses for successive 4-bit groups of said 16-byte
block, and transferring each of said 4-bit groups of said 16-byte block through a
4-bit data bus to the various pre-determined address locations in one of said 4xn
DRAM buffer memory configurations determined by said translating step.
2. A method according to claim 1 characterised by the steps of: providing a 8-bit parity
word for said 16-byte block, selecting a parity word row address, selecting a parity
word base column address for a first 4-bit group of said 8-bit parity word, incrementing
said parity word column address by four to provide a second column address for a second
4-bit group of said 8-bit parity word, and transferring said 8-bit parity word through
said 4-bit data bus to the various predetermined address locations for said 8-bit
parity word.
3. A method according to claim 1 or 2 characterised in that the translating step includes
the step of selecting the row and column addresses using multiplexer circuits.
4. A method according to claim 1 or 2 characterised by the step of latching the row and
column addresses into latch circuits, the incrementing step including the step of
feeding outputs from the latch circuits through multiplexing circuits connected to
the inputs of the latch circuits to increment the column addresses.
5. A method according to any of claims 1 to 4 characterised in that the 4xn DRAM buffer
memory comprises a DRAM operating in a page mode.
6. A method according to any of claims 1 to 4 characterised in that the 4xn DRAM buffer
memory comprises a DRAM operating in a static column mode.
7. A method according to claim 2 characterised in that the step of transferring the 8-bit
parity word includes the step of forming an address for said 8-bit parity word by
right-shifting the base column address of the 16-byte block by a pre-determined number
of bits and filling the vacated upper bit positions with 1-bits.